# AgentSky Pricing

Canonical URL: https://agentsky.dev/pricing

## Summary

- **Model usage**: per-token at the published rates listed below. Claude, GPT and DeepSeek bill at 30% off the maker's list price; every other model at the maker's price plus 10%.
- **Harness**: no charge of its own — every one of the 8 agents (Claude Code, Codex, OpenClaw, opencode, Hermes, Kimi Code, DeepSeek Harness, pi) runs at no harness charge; you pay only for the model, compute, capabilities, and connectors it uses.
- **Image, video & audio models**: billed per asset or per second of media when an agent calls them, at the maker's cost plus a 10% margin (rates below).
- **Paid tools & connectors**: billed only when used, at the rates listed below.
- **Compute**: per-second while the agent is awake, by machine shape (rates below); $0 while parked.
- **Storage**: 1 GiB per workspace free; $0.04 per GiB-month above that, from December 12, 2026. Measured and billed daily, a month counted as 30 days; each workspace's first 30 days of storage are free. Files are kept even while parked and are never deleted for an unpaid balance.
- **BYO subscription**: connect a Claude Pro/Max, ChatGPT, Cursor, Grok or Muse plan and model usage on eligible agents and models bills to your subscription, not your AgentSky balance. Models a plan doesn't serve use credits.
- **On your Mac**: in the AgentSky Mac app, Claude Code and Codex run on your Mac with your own Claude or ChatGPT plan and use no AgentSky credits (signed in with a personal API key, local runs use credits).
- **Workspace wallet**: a workspace shared with teammates can have its own wallet, funded by its owners and admins, with a monthly spending limit per member. Chats in that workspace are paid from it, whoever sends the message; a teammate who continues a chat synced from someone's Mac is paid for by the workspace wallet or their own balance, never by the Mac's owner.

## How pricing works

A session can include four billed components:

1. **Model tokens** — input and output tokens charged at the published per-model rates on the pricing page.
2. **Compute** — per-second while your agent is running; nothing when parked.
3. **Image, video & audio models and paid tools** — generation, transcription, search and browser actions, only when invoked.
4. **Connectors** — one charge for each request that reaches the connector provider.

Stored files bill per workspace, not per session: 1 GiB per workspace free; $0.04 per GiB-month above that, from December 12, 2026.

## Model rates

Input / output, USD per 1M tokens, as billed. Claude, GPT and DeepSeek bill 30% under the maker's published API price, which follows in parentheses; every other model bills at the maker's price plus 10%. Models from the wider catalog, served through OpenRouter, bill at their cheapest provider's price, including OpenRouter's fee, plus 10%, with no discount.

- **Claude Opus 5.5**: $2.80 in / $14.00 out (30% off the published $4.00 / $20.00)
- **Claude Sonnet 5.5**: $1.40 in / $7.00 out (30% off the published $2.00 / $10.00)
- **Claude Haiku 5.5**: $0.07 in / $0.35 out (30% off the published $0.10 / $0.50)
- **Claude Fable 5.1**: $7.00 in / $35.00 out (30% off the published $10.00 / $50.00)
- **Claude Opus 5**: $3.50 in / $17.50 out (30% off the published $5.00 / $25.00)
- **Claude Sonnet 5**: $1.40 in / $7.00 out (30% off the published $2.00 / $10.00)
- **Claude Haiku 4.5**: $0.70 in / $3.50 out (30% off the published $1.00 / $5.00)
- **GPT-6.1 Sol**: $1.40 in / $7.00 out (30% off the published $2.00 / $10.00)
- **GPT-6 Sol**: $1.40 in / $7.00 out (30% off the published $2.00 / $10.00)
- **GPT-6 Luna**: $0.07 in / $0.35 out (30% off the published $0.10 / $0.50)
- **GPT-5.6 Sol**: $3.50 in / $21.00 out (30% off the published $5.00 / $30.00)
- **GPT-5.6 Terra**: $1.40 in / $8.40 out (30% off the published $2.00 / $12.00)
- **GPT-5.6 Luna**: $0.14 in / $0.84 out (30% off the published $0.20 / $1.20)
- **GPT-6 Astra**: $7.00 in / $35.00 out (30% off the published $10.00 / $50.00)
- **DeepSeek V4 Pro**: $0.924 in / $2.772 out (30% off the published $1.32 / $3.96)
- **DeepSeek V4.1 Flash**: $0.21 in / $0.84 out (30% off the published $0.30 / $1.20)
- **Gemini 4 Argon**: $4.40 in / $22.00 out
- **Gemini 3.5 Flash-Lite**: $0.33 in / $2.75 out
- **Gemini 3.8 Flash**: $1.65 in / $8.25 out
- **GLM-5.3**: $1.54 in / $4.84 out
- **GLM-5.3-Flash**: $0.165 in / $0.55 out
- **Kimi K3.1 Pro**: $2.20 in / $11.00 out
- **Kimi K3**: $3.30 in / $16.50 out
- **Kimi K2.7 Code**: $1.045 in / $4.40 out
- **Grok 4.6**: $2.20 in / $6.60 out
- **Grok 4.7**: $2.20 in / $6.60 out
- **MiniMax-M3**: $0.33 in / $1.32 out
- **MiniMax-M3.1-Flash**: $0.165 in / $0.66 out
- **Jev 1.13**: $0.0462 in / $0.00 out
- **Muse Spark 1.1**: $1.375 in / $4.675 out
- **Muse Wonder 1.0**: $4.40 in / $22.00 out
- **Muse Spark 1.3**: $1.375 in / $4.675 out
- **Muse Spark 1.2**: $1.375 in / $4.675 out

## Compute rates

Billed per second while the agent is awake. $0 while suspended or parked.

- **1 vCPU · 512 MB**: $0.117/hour awake, $84.24/month always on
- **1 vCPU · 1 GB**: $0.133/hour awake, $95.90/month always on
- **1 vCPU · 2 GB**: $0.166/hour awake, $119.23/month always on
- **1 vCPU · 4 GB**: $0.230/hour awake, $165.89/month always on
- **2 vCPU · 512 MB**: $0.218/hour awake, $156.82/month always on
- **2 vCPU · 1 GB**: $0.234/hour awake, $168.48/month always on
- **2 vCPU · 2 GB**: $0.266/hour awake, $191.81/month always on
- **2 vCPU · 4 GB**: $0.331/hour awake, $238.46/month always on
- **2 vCPU · 8 GB** (default): $0.461/hour awake, $331.78/month always on
- **4 vCPU · 1 GB**: $0.436/hour awake, $313.63/month always on
- **4 vCPU · 2 GB**: $0.468/hour awake, $336.96/month always on
- **4 vCPU · 4 GB**: $0.533/hour awake, $383.62/month always on
- **4 vCPU · 8 GB**: $0.662/hour awake, $476.93/month always on
- **8 vCPU · 2 GB**: $0.871/hour awake, $627.26/month always on
- **8 vCPU · 4 GB**: $0.936/hour awake, $673.92/month always on
- **8 vCPU · 8 GB**: $1.066/hour awake, $767.23/month always on

## Image, video & audio models

Billed per asset or per second of media at the maker's cost plus a 10% margin. The provider is the model's maker.

- **Image generation** (OpenAI): $0.2321/asset
- **Video generation** (ByteDance): $0.1672/second
- **Image-to-video** (Alibaba): $0.02684/second
- **H3 video generation** (MiniMax): $0.088–$0.143/second
- **Audio transcription** (Fish Audio): $0.00011/second

## Paid tools & connectors

- **Web search** (TinyFish): $0.01/run
- **Web content** (Exa): $0.002/page
- **Web fetch** (TinyFish): $0.004/page
- **Browser session** (TinyFish): $0.03/minute
- **Browser automation** (TinyFish): $0.032/step
- **Background removal** (rembg): $0.01/asset
- **Code Review** (Provider unavailable): $0/run
- **Connector call** (AgentSky): $0.029/call

Prepaid credits are available in increments from $10. Top-ups add Stripe's card processing fee (2.9% + $0.30, shown at checkout) — e.g. $100 in credits costs $103.30 and credits $100.00 in full. There are no seat fees, no minimum commitments, and no card required to start.

Full pricing page with interactive estimator: https://agentsky.dev/pricing